| Digital image/video segmentation and matting is one of the most fundamental and challenging problems in computer vision and graphics. One popular application, as in many movie special effects, is to extract the foreground object out of the background scene, and composite it onto a new background with minimal visual artifacts. In my dissertation, I will propose several efficient object cutout systems with the state-of-the-art performance.;The first system is an interactive framework for soft segmentation and matting of natural images/videos. The proposed technique is based on the optimal linear time, computation of weighted geodesic distances to user-specified scribbles. The weights are based on spatial and/or temporal gradients, considering the statistics of the pixels scribbled by the user, without explicit optical flow. An automatic localized refinement step follows in order to further improve the results and accurately compute the corresponding matte function. The framework is complemented with numerous and diverse examples and comparisons with recent literature.;The second system, Video SnapCut1, was designed to meet the standards of the professional video production community. In this system segmentation is achieved by the collaboration of a set of local classifiers that adaptively integrate multiple local features. By localizing the classifiers, our system achieves significantly better results than previous systems for complicated videos, including dynamic background and non-rigid foreground deformations. We further show how this segmentation paradigm naturally supports local user editing and propagates them across time. The object cutout system is completed with a novel coherent matting technique. A comprehensive testing and comparison with previous techniques is presented to demonstrate the efficiency and robustness of the system.;In the third part I will present Dynamic Color Flow2, which unlike previous approaches, incorporates motion estimation into color modeling in a probabilistic framework, and adaptively changes model parameters to match the local properties of the motion. I will show how to apply this color model to both foreground and background layers in a balanced way for efficient object segmentation in video. Experimental results show that when compared with previous approaches, our model produces more accurate video object segmentation results.;1This work was jointly developed by the author and Jue Wang, David Simons, Adobe Systems. 2Jointly developed by the author and Jue Wang, Adobe Systems. |
